Why is JITNA called the 'HTTP of Agentic AI'?

HTTP defines how web browsers and servers communicate — regardless of who built them. JITNA defines how AI agents communicate tasks and memory across any model and any platform. It's a protocol layer, not merely a prompt format.

Does JITNA replace prompt engineering?

Yes, structurally. Prompt engineering relies on model-specific phrasing that breaks when models change. JITNA uses structured primitives that automatically translate into optimal prompt format for each model in the roster — humans declare intent, models receive optimally-formatted instructions.

G3 — The Language

JITNA Genome

The universal 6-primitive language for agentic AI — the HTTP of the AI era.

"Prompt Engineering won't last. Protocols do."

JITNA RFC-001 — Open Protocol Specification, Apache 2.0

6 Primitives: I / D / Δ / A / R / M

Intent

The core 'what': a precise, machine-readable statement of what the agent is asked to achieve. Intent is not a prompt — it is a structured declaration that survives model version upgrades and API changes.

I: Summarise Q4 financial report into 5 executive bullets

Data

The inputs: files, database references, API endpoints, prior conversation context. Data declarations are explicit — no hidden state, no ambient context. What you declare is what the agent uses.

D: file=reports/q4_2025.pdf, context=prior_summary_id:8821

Delta

The change specification: what should be different after this task completes vs. before. Delta is the success condition expressed as a differential — not 'write a report' but 'the report exists and contains bullets 1–5'. Without Delta, verification has no target.

Δ: output=5_bullets, format=markdown, word_limit=30_per_bullet

Action/Approach

The execution constraints and human authority parameter. A=0 means zero output regardless of model confidence — the absolute constitutional kill switch. A=1 to A=5 define increasingly autonomous execution modes. This is how FDIA's F=(D^I)×A operationalises human sovereignty over AI.

A: autonomy=3, model_tier=hexacore, max_cost=0.50usd

Reflection

The post-execution self-assessment. Every completed task generates an R block: what worked, what degraded, what was uncertain. R blocks feed the WF-05 Reflector phase and are the raw material for JITNA's continuous self-improvement.

R: accuracy_confidence=0.91, ambiguity_detected=false, latency=1.3s

Memory

The persistence specification: what should be written to the RCT Knowledge Vault (long-term), session state (medium-term), or discarded. M declarations make memory explicit — agents cannot silently accumulate state without the human's declared permission.

M: persist=summary, scope=project:q4, ttl=90days, vault=user_private

3 Syntax Forms

Minimal

I + A only — fastest path, uses system defaults for D/Δ/R/M

Use for: Quick queries, low-stakes tasks

Standard

I + D + Δ + A — explicit inputs and success conditions, default memory

Use for: Business tasks, report generation

Full

All 6 primitives declared — complete audit trail, full memory control, explicit reflection

Use for: Enterprise tasks, multi-step automation

ARTENT Genome

RCT Codex Genome

The Constitution — 10 Codices